Extract final uncertainty mask for each frame

[NnAracy]

Hi, thank you for the great work. I'm a beginner to this, can I get any hint from you of how to get the frame-wise uncertainty masks after all the refinements are done? In what format will the extracted masks be?

[Jianhao-zheng]

Hi, thank you for your interest in our work. There will be uncertainty_mlp_weight.pth saved in {WIldGS-SLAM dir}/output/{dataset}/{sequence}, which stores the checkpoing of online-trained uncertainty MLP. If you want to get the uncertainty mask, you will need to load that checkpoings and run img -> DinoV2 (pass to the uncertainty MLP)-> uncertainty map

You can also check this function to quickly understand the above process (

WildGS-SLAM/src/mapper.py

Line 1428 in 498c768

def get_viewpoint_uncertainty_no_grad(self, viewpoint: Camera) -> torch.Tensor:

).